com.github.asus4.onnxruntime 0.3.3 → 0.3.4

package/Plugins/{iOS~/onnxruntime.xcframework/ios-arm64/onnxruntime.framework → iOS/onnxruntime.xcframework/ios-arm64_x86_64-maccatalyst/onnxruntime.framework/Versions/A}/Headers/onnxruntime_session_options_config_keys.h

@@ -250,6 +250,51 @@ static const char* const kOrtSessionOptionsOptimizedModelExternalInitializersFil
 static const char* const kOrtSessionOptionsOptimizedModelExternalInitializersMinSizeInBytes =
     "session.optimized_model_external_initializers_min_size_in_bytes";
 
+// When loading model from memory buffer and the model has external initializers
+// Use this config to set the external data file folder path
+// All external data files should be in the same folder
+static const char* const kOrtSessionOptionsModelExternalInitializersFileFolderPath =
+    "session.model_external_initializers_file_folder_path";
+
+// Use this config when saving pre-packed constant initializers to an external data file.
+// This allows you to memory map pre-packed initializers on model load and leave it to
+// to the OS the amount of memory consumed by the pre-packed initializers. Otherwise,
+// pre-packed data resides on the heap.
+//
+// - "0": Default is not save pre-packed initializers to a data file.
+// - "1": Save pre-packed constant initializers to an external data file.
+// Sample usage: sess_options.add_session_config_entry(kOrtSessionOptionsSavePrePackedConstantInitializers,  "1")
+static const char* const kOrtSessionOptionsSavePrePackedConstantInitializers =
+    "session.save_external_prepacked_constant_initializers";
+
+// Use this config when you want to collect memory stats for each node in the graph.
+// The file format is a CSV file with the following columns:
+// The file will be created if it does not exist, and will be overwritten if it does.
+//
+// The content of the file can be used to estimate memory requirements at run time including
+// the temporary allocations. This operation is preferably done on a CPU device, as the model may exceed
+// device memory limits in constrained environments. When enabling this option, it is important to disable
+// memory patterns, as they tend to allocate large blocks to avoid fragmentation and accommodate needs of multiple
+// kernels. Memory patterns may make it difficult to allocate on a device with limited memory.
+//
+// The collected stats then can be used to partition the graph among the devices in a way that only the
+// required memory is allocated on each device.
+//
+// node_name, initializers_memory, dynamic_outputs_sizes, temp_allocations_size
+//
+// - "full path to file": there is not a default for this option. If the file can not be opened for writing, an error will be returned.
+static const char* const kOrtSessionOptionsCollectNodeMemoryStatsToFile = "session.collect_node_memory_stats_to_file";
+
+/// This is a composite CSV setting formatted as "memory limit in kb,file name for collected stats"
+/// "limit > 0": enables Capacity Aware Partitioning for Cuda EP. `limit` is optional and when absent
+/// the provider may attempt to figure out the memory available automatically.
+/// The setting with no limit is expected to look like: ",file name for collected stats"
+///  The EP will place nodes on device "file name" :
+/// this file is expected to be found at the same folder with the model. The file contains
+/// pre-recorded stats collected when running with kOrtSessionOptionsCollectNodeMemoryStatsToFile enforce (see above)
+static const char* const kOrtSessionOptionsResourceCudaPartitioningSettings =
+    "session.resource_cuda_partitioning_settings";
+
 // Enable EP context feature to dump the partitioned graph which includes the EP context into Onnx file.
 // The dumped Onnx model with EP context can be used for future inference to avoid the EP graph partitioning/compile overhead.
 // "0": disable. (default)
@@ -258,11 +303,12 @@ static const char* const kOrtSessionOptionEpContextEnable = "ep.context_enable";
 
 // Specify the file path for the Onnx model which has EP context.
 // Default to original_file_name_ctx.onnx if not specified
+// Folder is not a valid option
 static const char* const kOrtSessionOptionEpContextFilePath = "ep.context_file_path";
 
 // Flag to specify whether to dump the EP context into the Onnx model.
-// "0": dump the EP context into separate file, keep the file name in the Onnx model.
-// "1": dump the EP context into the Onnx model. (default).
+// "0": dump the EP context into separate file, keep the file name in the Onnx model. (default).
+// "1": dump the EP context into the Onnx model.
 static const char* const kOrtSessionOptionEpContextEmbedMode = "ep.context_embed_mode";
 
 // Specify the EPContext node name prefix to make it unique
@@ -272,6 +318,11 @@ static const char* const kOrtSessionOptionEpContextNodeNamePrefix = "ep.context_
 // Share EP related resources across EPs
 static const char* const kOrtSessionOptionShareEpContexts = "ep.share_ep_contexts";
 
+// Use this config when dumping EP context model with an external initializers file
+// All initializers will be inside the external data file if specified, otherwise all in Onnx file
+static const char* const kOrtSessionOptionsEpContextModelExternalInitializersFileName =
+    "ep.context_model_external_initializers_file_name";
+
 // Gemm fastmath mode provides fp32 gemm acceleration with bfloat16 based matmul.
 // Option values:
 // - "0": Gemm FastMath mode is not enabled. [DEFAULT]

package/Plugins/{iOS~/onnxruntime.xcframework/ios-arm64_x86_64-simulator/onnxruntime.framework → iOS/onnxruntime.xcframework/ios-arm64_x86_64-maccatalyst/onnxruntime.framework/Versions/A/Resources}/Info.plist

@@ -9,9 +9,9 @@
 	<key>CFBundleIdentifier</key>
 	<string>com.microsoft.onnxruntime</string>
 	<key>CFBundleVersion</key>
-	<string>1.20.0</string>
+	<string>1.21.0</string>
 	<key>CFBundleShortVersionString</key>
-	<string>1.20.0</string>
+	<string>1.21.0</string>
 	<key>CFBundleSignature</key>
 	<string>????</string>
 	<key>CFBundlePackageType</key>

package/Plugins/iOS/onnxruntime.xcframework/ios-arm64_x86_64-simulator/onnxruntime.framework/Headers/coreml_provider_factory.h

@@ -0,0 +1,90 @@
+// Copyright (c) Microsoft Corporation. All rights reserved.
+// Licensed under the MIT License.
+#pragma once
+
+#include "onnxruntime_c_api.h"
+
+// COREMLFlags are bool options we want to set for CoreML EP
+// This enum is defined as bit flags, and cannot have negative value
+// To generate an uint32_t coreml_flags for using with OrtSessionOptionsAppendExecutionProvider_CoreML below,
+//   uint32_t coreml_flags = 0;
+//   coreml_flags |= COREML_FLAG_USE_CPU_ONLY;
+enum COREMLFlags {
+  COREML_FLAG_USE_NONE = 0x000,
+
+  // Using CPU only in CoreML EP, this may decrease the perf but will provide
+  // reference output value without precision loss, which is useful for validation
+  COREML_FLAG_USE_CPU_ONLY = 0x001,
+
+  // Enable CoreML EP on subgraph
+  COREML_FLAG_ENABLE_ON_SUBGRAPH = 0x002,
+
+  // By default CoreML Execution provider will be enabled for all compatible Apple devices
+  // Enable this option will only enable CoreML EP for Apple devices with ANE (Apple Neural Engine)
+  // Please note, enable this option does not guarantee the entire model to be executed using ANE only
+  COREML_FLAG_ONLY_ENABLE_DEVICE_WITH_ANE = 0x004,
+
+  // Only allow CoreML EP to take nodes with inputs with static shapes. By default it will also allow inputs with
+  // dynamic shapes. However, the performance may be negatively impacted if inputs have dynamic shapes.
+  COREML_FLAG_ONLY_ALLOW_STATIC_INPUT_SHAPES = 0x008,
+
+  // Create an MLProgram. By default it will create a NeuralNetwork model. Requires Core ML 5 or later.
+  COREML_FLAG_CREATE_MLPROGRAM = 0x010,
+
+  // https://developer.apple.com/documentation/coreml/mlcomputeunits?language=objc
+  // there are four compute units:
+  // MLComputeUnitsCPUAndNeuralEngine|MLComputeUnitsCPUAndGPU|MLComputeUnitsCPUOnly|MLComputeUnitsAll
+  // different CU will have different performance and power consumption
+  COREML_FLAG_USE_CPU_AND_GPU = 0x020,
+  // Keep COREML_FLAG_LAST at the end of the enum definition
+  // And assign the last COREMLFlag to it
+  COREML_FLAG_LAST = COREML_FLAG_USE_CPU_AND_GPU,
+};
+
+// MLComputeUnits can be one of the following values:
+// 'MLComputeUnitsCPUAndNeuralEngine|MLComputeUnitsCPUAndGPU|MLComputeUnitsCPUOnly|MLComputeUnitsAll'
+// these values are intended to be used with Ort::SessionOptions::AppendExecutionProvider (C++ API)
+// and SessionOptionsAppendExecutionProvider (C API). For the old API, use COREMLFlags instead.
+static const char* const kCoremlProviderOption_MLComputeUnits = "MLComputeUnits";
+static const char* const kCoremlProviderOption_ModelFormat = "ModelFormat";
+// same as COREML_FLAG_ONLY_ALLOW_STATIC_INPUT_SHAPES
+static const char* const kCoremlProviderOption_RequireStaticInputShapes = "RequireStaticInputShapes";
+static const char* const kCoremlProviderOption_EnableOnSubgraphs = "EnableOnSubgraphs";
+// provided by https://developer.apple.com/documentation/coreml/mloptimizationhints-swift.struct/specializationstrategy-swift.property
+// Core ML segments the model’s compute graph and specializes each segment for the target compute device.
+// This process can affect the model loading time and the prediction latency.
+// Use this option to tailor the specialization strategy for your model.
+static const char* const kCoremlProviderOption_SpecializationStrategy = "SpecializationStrategy";
+// Profile the Core ML MLComputePlan.
+// This logs the hardware each operator is dispatched to and the estimated execution time.
+// Intended for developer usage but provide useful diagnostic information if performance is not as expected.
+static const char* const kCoremlProviderOption_ProfileComputePlan = "ProfileComputePlan";
+// please refer to https://developer.apple.com/documentation/coreml/mlmodelconfiguration/allowlowprecisionaccumulationongpu
+static const char* const kCoremlProviderOption_AllowLowPrecisionAccumulationOnGPU = "AllowLowPrecisionAccumulationOnGPU";
+// Specify the directory to cache any CoreML models created from the ONNX model in.
+// CoreML EP will convert onnx subgraph to CoreML model and save to disk.
+// If this path is not specified, the model will be saved to a temp directory and deleted after the session is closed.
+// otherwise, the model will be saved to the specified path and User should manage to delete the model.
+
+// we do NOT detect if the onnx model has changed and no longer matches the cached model.
+// the user should carefully manage the cache if modifying/replacing a model.
+// The cache key is generated by
+//    1. User provided key in metadata_props if found (preferred)
+//    2. Hash of the model url the inference session was created with
+//    3. Hash of the input/output names of the model
+// Please find out how to set metadata_props in the onnxruntime API documentation. https://onnxruntime.ai/docs/execution-providers/CoreML-ExecutionProvider.html#configuration-options
+static const char* const kCoremlProviderOption_ModelCacheDirectory = "ModelCacheDirectory";
+
+// User provided cache-key in metadata_props.
+static const char* const kCOREML_CACHE_KEY = "COREML_CACHE_KEY";
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+ORT_EXPORT ORT_API_STATUS(OrtSessionOptionsAppendExecutionProvider_CoreML,
+                          _In_ OrtSessionOptions* options, uint32_t coreml_flags);
+
+#ifdef __cplusplus
+}
+#endif

package/Plugins/{iOS~ → iOS}/onnxruntime.xcframework/ios-arm64_x86_64-simulator/onnxruntime.framework/Headers/onnxruntime_c_api.h

@@ -38,7 +38,7 @@
  *
  * This value is used by some API functions to behave as this version of the header expects.
  */
-#define ORT_API_VERSION 20
+#define ORT_API_VERSION 21
 
 #ifdef __cplusplus
 extern "C" {
@@ -46,7 +46,7 @@ extern "C" {
 
 //! @}
 // SAL2 Definitions
-#ifndef _WIN32
+#ifndef _MSC_VER
 #define _In_
 #define _In_z_
 #define _In_opt_
@@ -626,8 +626,13 @@ typedef struct OrtMIGraphXProviderOptions {
 } OrtMIGraphXProviderOptions;
 
 /** \brief OpenVINO Provider Options
- *
- * \see OrtApi::SessionOptionsAppendExecutionProvider_OpenVINO
+ *  \brief This Struct is frozen since ORT 1.13.0. Its maintained part of Legacy API for compatibility.
+ *  \brief For latest OpenVINO Provider Options update to the ProviderOptions map.
+ *  \brief Latest OpenVINO Provider Options are listed in the
+ *  \htmlonly
+ *  <a href="https://onnxruntime.ai/docs/execution-providers/OpenVINO-ExecutionProvider.html#summary-of-options">onnxruntime document.</a>
+ *  \endhtmlonly
+ * \see OrtApi::SessionOptionsAppendExecutionProvider()
  */
 typedef struct OrtOpenVINOProviderOptions {
 #ifdef __cplusplus
@@ -645,7 +650,7 @@ typedef struct OrtOpenVINOProviderOptions {
    * Valid settings are one of: "CPU_FP32", "CPU_FP16", "GPU_FP32", "GPU_FP16"
    */
   const char* device_type;
-  unsigned char enable_npu_fast_compile;
+  unsigned char enable_npu_fast_compile;  ///< 0 = disabled, nonzero = enabled
   const char* device_id;
   size_t num_of_threads;  ///< 0 = Use default number of threads
   const char* cache_dir;  // path is set to empty by default
@@ -3660,8 +3665,19 @@ struct OrtApi {
    *     - "1": Enabled.
    *   "offload_graph_io_quantization": Offload graph input quantization and graph output dequantization to another
    *   execution provider (typically CPU EP).
-   *     - "0": Default. Disabled. QNN EP will handle quantization and dequantization of graph I/O.
+   *     - "0": Disabled. QNN EP will handle quantization and dequantization of graph I/O.
+   *     - "1": Enabled. This is the default value.
+   *   "enable_htp_spill_fill_buffer": Enable HTP spill fill buffer setting. The flag is used while generating context binary.
+   *     - "0": Default. Disabled.
+   *     - "1": Enabled.
+   *   "enable_htp_shared_memory_allocator": Enable the QNN HTP shared memory allocator. Requires libcdsprpc.so/dll to
+   *   be available.
+   *     - "0": Default. Disabled.
    *     - "1": Enabled.
+   *   "dump_json_qnn_graph": Set to "1" to dump QNN graphs generated by QNN EP as JSON files. Each graph partition
+   *      assigned to QNN EP is dumped to a separate file.
+   *   "json_qnn_graph_dir": Directory in which to dump QNN JSON graphs. If not specified, QNN graphs are dumped in the
+   *      program's current working directory. Ignored if "dump_json_qnn_graph" is not set.
    *
    * SNPE supported keys:
    *   "runtime": SNPE runtime engine, options: "CPU", "CPU_FLOAT32", "GPU", "GPU_FLOAT32_16_HYBRID", "GPU_FLOAT16",
@@ -4607,6 +4623,8 @@ struct OrtApi {
    * \param[in] num_keys
    *
    * \snippet{doc} snippets.dox OrtStatus Return Value
+   *
+   * \since Version 1.17.
    */
   ORT_API2_STATUS(SessionOptionsAppendExecutionProvider_OpenVINO_V2,
                   _In_ OrtSessionOptions* options,
@@ -4624,6 +4642,8 @@ struct OrtApi {
    * \param[in] num_keys
    *
    * \snippet{doc} snippets.dox OrtStatus Return Value
+   *
+   * \since Version 1.18.
    */
   ORT_API2_STATUS(SessionOptionsAppendExecutionProvider_VitisAI,
                   _In_ OrtSessionOptions* options,
@@ -4637,7 +4657,10 @@ struct OrtApi {
    *  \param[in] mem_info OrtMemoryInfo instance
    *  \param[in] count_or_bytes How many bytes is this scratch buffer
    *  \param[out] out A pointer to the scrach buffer
+   *
    *  \snippet{doc} snippets.dox OrtStatus Return Value
+   *
+   * \since Version 1.18.
    */
   ORT_API2_STATUS(KernelContext_GetScratchBuffer, _In_ const OrtKernelContext* context, _In_ const OrtMemoryInfo* mem_info, _In_ size_t count_or_bytes, _Outptr_ void** out);
 
@@ -4648,6 +4671,8 @@ struct OrtApi {
    * \param[out] out A pointer to OrtAllocator
    *
    * \snippet{doc} snippets.dox OrtStatus Return Value
+   *
+   * \since Version 1.18.
    */
   ORT_API2_STATUS(KernelInfoGetAllocator, _In_ const OrtKernelInfo* info, _In_ OrtMemType mem_type, _Outptr_ OrtAllocator** out);
 
@@ -4669,6 +4694,8 @@ struct OrtApi {
    * \param[in] num_external_initializer_files Number of external files
    *
    * \snippet{doc} snippets.dox OrtStatus Return Value
+   *
+   * \since Version 1.18.
    */
   ORT_API2_STATUS(AddExternalInitializersFromFilesInMemory, _In_ OrtSessionOptions* options,
                   _In_reads_(num_external_initializer_files) const ORTCHAR_T* const* external_initializer_file_names,
@@ -4691,6 +4718,8 @@ struct OrtApi {
    *                  OrtApi::ReleaseLoraAdapter.
    *
    * \snippet{doc} snippets.dox OrtStatus Return Value
+   *
+   * \since Version 1.20.
    */
   ORT_API2_STATUS(CreateLoraAdapter, const ORTCHAR_T* adapter_file_path, _In_ OrtAllocator* allocator,
                   _Outptr_ OrtLoraAdapter** out);
@@ -4709,6 +4738,8 @@ struct OrtApi {
    *                  OrtApi::ReleaseLoraAdapter.
    *
    * \snippet{doc} snippets.dox OrtStatus Return Value
+   *
+   * \since Version 1.20.
    */
   ORT_API2_STATUS(CreateLoraAdapterFromArray, _In_ const void* bytes, size_t num_bytes, _In_ OrtAllocator* allocator,
                   _Outptr_ OrtLoraAdapter** out);
@@ -4730,6 +4761,8 @@ struct OrtApi {
    * \param[in] adapter OrtLoraAdapter instance
    *
    * \snippet{doc} snippets.dox OrtStatus Return Value
+   *
+   * \since Version 1.20.
    */
   ORT_API2_STATUS(RunOptionsAddActiveLoraAdapter, _Inout_ OrtRunOptions* options, _In_ const OrtLoraAdapter* adapter);
 
@@ -4748,6 +4781,8 @@ struct OrtApi {
    * \param[in] kv_len Number of elements in the keys and values arrays
    *
    * \snippet{doc} snippets.dox OrtStatus Return Value
+   *
+   * \since Version 1.20.
    */
   ORT_API2_STATUS(SetEpDynamicOptions, _Inout_ OrtSession* sess, _In_reads_(kv_len) const char* const* keys,
                   _In_reads_(kv_len) const char* const* values, _In_ size_t kv_len);

package/Plugins/{iOS~/onnxruntime.xcframework/macos-arm64_x86_64/onnxruntime.framework/Versions/A → iOS/onnxruntime.xcframework/ios-arm64_x86_64-simulator/onnxruntime.framework}/Headers/onnxruntime_cxx_api.h

@@ -650,6 +650,9 @@ using AllocatedStringPtr = std::unique_ptr<char, detail::AllocatedFree>;
  *  constructors to construct an instance of a Status object from exceptions.
  */
 struct Status : detail::Base<OrtStatus> {
+  using Base = detail::Base<OrtStatus>;
+  using Base::Base;
+
   explicit Status(std::nullptr_t) noexcept {}               ///< Create an empty object, must be assigned a valid one to be used
   explicit Status(OrtStatus* status) noexcept;              ///< Takes ownership of OrtStatus instance returned from the C API.
   explicit Status(const Exception&) noexcept;               ///< Creates status instance out of exception
@@ -728,6 +731,9 @@ struct Env : detail::Base<OrtEnv> {
  *
  */
 struct CustomOpDomain : detail::Base<OrtCustomOpDomain> {
+  using Base = detail::Base<OrtCustomOpDomain>;
+  using Base::Base;
+
   explicit CustomOpDomain(std::nullptr_t) {}  ///< Create an empty CustomOpDomain object, must be assigned a valid one to be used
 
   /// \brief Wraps OrtApi::CreateCustomOpDomain
@@ -963,8 +969,10 @@ struct SessionOptions : detail::SessionOptionsImpl<OrtSessionOptions> {
  *
  */
 struct ModelMetadata : detail::Base<OrtModelMetadata> {
-  explicit ModelMetadata(std::nullptr_t) {}                                   ///< Create an empty ModelMetadata object, must be assigned a valid one to be used
-  explicit ModelMetadata(OrtModelMetadata* p) : Base<OrtModelMetadata>{p} {}  ///< Used for interop with the C API
+  using Base = detail::Base<OrtModelMetadata>;
+  using Base::Base;
+
+  explicit ModelMetadata(std::nullptr_t) {}  ///< Create an empty ModelMetadata object, must be assigned a valid one to be used
 
   /** \brief Returns a copy of the producer name.
    *
@@ -1237,6 +1245,9 @@ using ConstTensorTypeAndShapeInfo = detail::TensorTypeAndShapeInfoImpl<detail::U
  *
  */
 struct TensorTypeAndShapeInfo : detail::TensorTypeAndShapeInfoImpl<OrtTensorTypeAndShapeInfo> {
+  using Base = detail::TensorTypeAndShapeInfoImpl<OrtTensorTypeAndShapeInfo>;
+  using Base::Base;
+
   explicit TensorTypeAndShapeInfo(std::nullptr_t) {}                                                ///< Create an empty TensorTypeAndShapeInfo object, must be assigned a valid one to be used
   explicit TensorTypeAndShapeInfo(OrtTensorTypeAndShapeInfo* p) : TensorTypeAndShapeInfoImpl{p} {}  ///< Used for interop with the C API
   ConstTensorTypeAndShapeInfo GetConst() const { return ConstTensorTypeAndShapeInfo{this->p_}; }
@@ -1258,6 +1269,9 @@ using ConstSequenceTypeInfo = detail::SequenceTypeInfoImpl<detail::Unowned<const
  *
  */
 struct SequenceTypeInfo : detail::SequenceTypeInfoImpl<OrtSequenceTypeInfo> {
+  using Base = detail::SequenceTypeInfoImpl<OrtSequenceTypeInfo>;
+  using Base::Base;
+
   explicit SequenceTypeInfo(std::nullptr_t) {}                                                         ///< Create an empty SequenceTypeInfo object, must be assigned a valid one to be used
   explicit SequenceTypeInfo(OrtSequenceTypeInfo* p) : SequenceTypeInfoImpl<OrtSequenceTypeInfo>{p} {}  ///< Used for interop with the C API
   ConstSequenceTypeInfo GetConst() const { return ConstSequenceTypeInfo{this->p_}; }
@@ -1293,6 +1307,9 @@ using ConstMapTypeInfo = detail::MapTypeInfoImpl<detail::Unowned<const OrtMapTyp
  *
  */
 struct MapTypeInfo : detail::MapTypeInfoImpl<OrtMapTypeInfo> {
+  using Base = detail::MapTypeInfoImpl<OrtMapTypeInfo>;
+  using Base::Base;
+
   explicit MapTypeInfo(std::nullptr_t) {}                                          ///< Create an empty MapTypeInfo object, must be assigned a valid one to be used
   explicit MapTypeInfo(OrtMapTypeInfo* p) : MapTypeInfoImpl<OrtMapTypeInfo>{p} {}  ///< Used for interop with the C API
   ConstMapTypeInfo GetConst() const { return ConstMapTypeInfo{this->p_}; }
@@ -1324,6 +1341,9 @@ using ConstTypeInfo = detail::TypeInfoImpl<detail::Unowned<const OrtTypeInfo>>;
 /// the information about contained sequence or map depending on the ONNXType.
 /// </summary>
 struct TypeInfo : detail::TypeInfoImpl<OrtTypeInfo> {
+  using Base = detail::TypeInfoImpl<OrtTypeInfo>;
+  using Base::Base;
+
   explicit TypeInfo(std::nullptr_t) {}                                 ///< Create an empty TypeInfo object, must be assigned a valid one to be used
   explicit TypeInfo(OrtTypeInfo* p) : TypeInfoImpl<OrtTypeInfo>{p} {}  ///< C API Interop
 
@@ -1661,11 +1681,11 @@ using UnownedValue = detail::ValueImpl<detail::Unowned<OrtValue>>;
  */
 struct Value : detail::ValueImpl<OrtValue> {
   using Base = detail::ValueImpl<OrtValue>;
+  using Base::Base;
   using OrtSparseValuesParam = detail::OrtSparseValuesParam;
   using Shape = detail::Shape;
 
-  explicit Value(std::nullptr_t) {}         ///< Create an empty Value object, must be assigned a valid one to be used
-  explicit Value(OrtValue* p) : Base{p} {}  ///< Used for interop with the C API
+  explicit Value(std::nullptr_t) {}  ///< Create an empty Value object, must be assigned a valid one to be used
   Value(Value&&) = default;
   Value& operator=(Value&&) = default;
 
@@ -1941,6 +1961,10 @@ struct ArenaCfg : detail::Base<OrtArenaCfg> {
 /// This struct provides life time management for custom op attribute
 /// </summary>
 struct OpAttr : detail::Base<OrtOpAttr> {
+  using Base = detail::Base<OrtOpAttr>;
+  using Base::Base;
+
+  explicit OpAttr(std::nullptr_t) {}
   OpAttr(const char* name, const void* data, int len, OrtOpAttrType type);
 };
 
@@ -2106,10 +2130,10 @@ struct KernelContext {
   explicit KernelContext(OrtKernelContext* context);
   size_t GetInputCount() const;
   size_t GetOutputCount() const;
-  // If input is optional and is not present, the method returns en empty ConstValue
+  // If input is optional and is not present, the method returns an empty ConstValue
   // which can be compared to nullptr.
   ConstValue GetInput(size_t index) const;
-  // If outout is optional and is not present, the method returns en empty UnownedValue
+  // If output is optional and is not present, the method returns an empty UnownedValue
   // which can be compared to nullptr.
   UnownedValue GetOutput(size_t index, const int64_t* dim_values, size_t dim_count) const;
   UnownedValue GetOutput(size_t index, const std::vector<int64_t>& dims) const;
@@ -2183,6 +2207,8 @@ using ConstKernelInfo = detail::KernelInfoImpl<detail::Unowned<const OrtKernelIn
 /// so it does not destroy the pointer the kernel does not own.
 /// </summary>
 struct KernelInfo : detail::KernelInfoImpl<OrtKernelInfo> {
+  using Base = detail::KernelInfoImpl<OrtKernelInfo>;
+  using Base::Base;
   explicit KernelInfo(std::nullptr_t) {}     ///< Create an empty instance to initialize later
   explicit KernelInfo(OrtKernelInfo* info);  ///< Take ownership of the instance
   ConstKernelInfo GetConst() const { return ConstKernelInfo{this->p_}; }
@@ -2192,6 +2218,9 @@ struct KernelInfo : detail::KernelInfoImpl<OrtKernelInfo> {
 /// Create and own custom defined operation.
 /// </summary>
 struct Op : detail::Base<OrtOp> {
+  using Base = detail::Base<OrtOp>;
+  using Base::Base;
+
   explicit Op(std::nullptr_t) {}  ///< Create an empty Operator object, must be assigned a valid one to be used
 
   explicit Op(OrtOp*);  ///< Take ownership of the OrtOp

package/Plugins/{iOS~ → iOS}/onnxruntime.xcframework/ios-arm64_x86_64-simulator/onnxruntime.framework/Headers/onnxruntime_cxx_inline.h

@@ -51,7 +51,7 @@ inline void ThrowOnError(const Status& st) {
   }
 }
 
-inline Status::Status(OrtStatus* status) noexcept : Base<OrtStatus>{status} {
+inline Status::Status(OrtStatus* status) noexcept : detail::Base<OrtStatus>{status} {
 }
 
 inline Status::Status(const std::exception& e) noexcept {
@@ -1908,7 +1908,7 @@ inline void attr_utils::GetAttrs(const OrtKernelInfo* p, const char* name, std::
 
 inline KernelInfo::KernelInfo(OrtKernelInfo* info) : detail::KernelInfoImpl<OrtKernelInfo>{info} {}
 
-inline Op::Op(OrtOp* p) : Base<OrtOp>(p) {}
+inline Op::Op(OrtOp* p) : detail::Base<OrtOp>(p) {}
 
 inline Op Op::Create(const OrtKernelInfo* info, const char* op_name, const char* domain, int version,
                      const char** type_constraint_names,

package/Plugins/{iOS~ → iOS}/onnxruntime.xcframework/ios-arm64_x86_64-simulator/onnxruntime.framework/Headers/onnxruntime_session_options_config_keys.h

@@ -250,6 +250,51 @@ static const char* const kOrtSessionOptionsOptimizedModelExternalInitializersFil
 static const char* const kOrtSessionOptionsOptimizedModelExternalInitializersMinSizeInBytes =
     "session.optimized_model_external_initializers_min_size_in_bytes";
 
+// When loading model from memory buffer and the model has external initializers
+// Use this config to set the external data file folder path
+// All external data files should be in the same folder
+static const char* const kOrtSessionOptionsModelExternalInitializersFileFolderPath =
+    "session.model_external_initializers_file_folder_path";
+
+// Use this config when saving pre-packed constant initializers to an external data file.
+// This allows you to memory map pre-packed initializers on model load and leave it to
+// to the OS the amount of memory consumed by the pre-packed initializers. Otherwise,
+// pre-packed data resides on the heap.
+//
+// - "0": Default is not save pre-packed initializers to a data file.
+// - "1": Save pre-packed constant initializers to an external data file.
+// Sample usage: sess_options.add_session_config_entry(kOrtSessionOptionsSavePrePackedConstantInitializers,  "1")
+static const char* const kOrtSessionOptionsSavePrePackedConstantInitializers =
+    "session.save_external_prepacked_constant_initializers";
+
+// Use this config when you want to collect memory stats for each node in the graph.
+// The file format is a CSV file with the following columns:
+// The file will be created if it does not exist, and will be overwritten if it does.
+//
+// The content of the file can be used to estimate memory requirements at run time including
+// the temporary allocations. This operation is preferably done on a CPU device, as the model may exceed
+// device memory limits in constrained environments. When enabling this option, it is important to disable
+// memory patterns, as they tend to allocate large blocks to avoid fragmentation and accommodate needs of multiple
+// kernels. Memory patterns may make it difficult to allocate on a device with limited memory.
+//
+// The collected stats then can be used to partition the graph among the devices in a way that only the
+// required memory is allocated on each device.
+//
+// node_name, initializers_memory, dynamic_outputs_sizes, temp_allocations_size
+//
+// - "full path to file": there is not a default for this option. If the file can not be opened for writing, an error will be returned.
+static const char* const kOrtSessionOptionsCollectNodeMemoryStatsToFile = "session.collect_node_memory_stats_to_file";
+
+/// This is a composite CSV setting formatted as "memory limit in kb,file name for collected stats"
+/// "limit > 0": enables Capacity Aware Partitioning for Cuda EP. `limit` is optional and when absent
+/// the provider may attempt to figure out the memory available automatically.
+/// The setting with no limit is expected to look like: ",file name for collected stats"
+///  The EP will place nodes on device "file name" :
+/// this file is expected to be found at the same folder with the model. The file contains
+/// pre-recorded stats collected when running with kOrtSessionOptionsCollectNodeMemoryStatsToFile enforce (see above)
+static const char* const kOrtSessionOptionsResourceCudaPartitioningSettings =
+    "session.resource_cuda_partitioning_settings";
+
 // Enable EP context feature to dump the partitioned graph which includes the EP context into Onnx file.
 // The dumped Onnx model with EP context can be used for future inference to avoid the EP graph partitioning/compile overhead.
 // "0": disable. (default)
@@ -258,11 +303,12 @@ static const char* const kOrtSessionOptionEpContextEnable = "ep.context_enable";
 
 // Specify the file path for the Onnx model which has EP context.
 // Default to original_file_name_ctx.onnx if not specified
+// Folder is not a valid option
 static const char* const kOrtSessionOptionEpContextFilePath = "ep.context_file_path";
 
 // Flag to specify whether to dump the EP context into the Onnx model.
-// "0": dump the EP context into separate file, keep the file name in the Onnx model.
-// "1": dump the EP context into the Onnx model. (default).
+// "0": dump the EP context into separate file, keep the file name in the Onnx model. (default).
+// "1": dump the EP context into the Onnx model.
 static const char* const kOrtSessionOptionEpContextEmbedMode = "ep.context_embed_mode";
 
 // Specify the EPContext node name prefix to make it unique
@@ -272,6 +318,11 @@ static const char* const kOrtSessionOptionEpContextNodeNamePrefix = "ep.context_
 // Share EP related resources across EPs
 static const char* const kOrtSessionOptionShareEpContexts = "ep.share_ep_contexts";
 
+// Use this config when dumping EP context model with an external initializers file
+// All initializers will be inside the external data file if specified, otherwise all in Onnx file
+static const char* const kOrtSessionOptionsEpContextModelExternalInitializersFileName =
+    "ep.context_model_external_initializers_file_name";
+
 // Gemm fastmath mode provides fp32 gemm acceleration with bfloat16 based matmul.
 // Option values:
 // - "0": Gemm FastMath mode is not enabled. [DEFAULT]

package/Plugins/{iOS~/onnxruntime.xcframework/macos-arm64_x86_64/onnxruntime.framework/Versions/A/Resources → iOS/onnxruntime.xcframework/ios-arm64_x86_64-simulator/onnxruntime.framework}/Info.plist

@@ -9,9 +9,9 @@
 	<key>CFBundleIdentifier</key>
 	<string>com.microsoft.onnxruntime</string>
 	<key>CFBundleVersion</key>
-	<string>1.20.0</string>
+	<string>1.21.0</string>
 	<key>CFBundleShortVersionString</key>
-	<string>1.20.0</string>
+	<string>1.21.0</string>
 	<key>CFBundleSignature</key>
 	<string>????</string>
 	<key>CFBundlePackageType</key>

package/Plugins/iOS/onnxruntime.xcframework.meta

@@ -0,0 +1,61 @@
+fileFormatVersion: 2
+guid: 61e9efa0c322a4a17869727d5dc86a50
+PluginImporter:
+  externalObjects: {}
+  serializedVersion: 3
+  iconMap: {}
+  executionOrder: {}
+  defineConstraints: []
+  isPreloaded: 0
+  isOverridable: 1
+  isExplicitlyReferenced: 0
+  validateReferences: 1
+  platformData:
+    Android:
+      enabled: 0
+      settings:
+        AndroidLibraryDependee: UnityLibrary
+        AndroidSharedLibraryType: Executable
+        CPU: ARMv7
+    Any:
+      enabled: 0
+      settings:
+        Exclude Android: 1
+        Exclude Editor: 1
+        Exclude Linux64: 1
+        Exclude OSXUniversal: 1
+        Exclude Win: 1
+        Exclude Win64: 1
+        Exclude iOS: 0
+    Editor:
+      enabled: 0
+      settings:
+        CPU: AnyCPU
+        DefaultValueInitialized: true
+        OS: AnyOS
+    Linux64:
+      enabled: 0
+      settings:
+        CPU: None
+    OSXUniversal:
+      enabled: 0
+      settings:
+        CPU: None
+    Win:
+      enabled: 0
+      settings:
+        CPU: None
+    Win64:
+      enabled: 0
+      settings:
+        CPU: None
+    iOS:
+      enabled: 1
+      settings:
+        AddToEmbeddedBinaries: false
+        CPU: AnyCPU
+        CompileFlags: 
+        FrameworkDependencies: MobileCoreServices;CoreML;ImageIO;
+  userData: 
+  assetBundleName: 
+  assetBundleVariant: 

package/Plugins/iOS.meta

@@ -0,0 +1,8 @@
+fileFormatVersion: 2
+guid: 59d968276c1924908bedb8d0ca6c447c
+folderAsset: yes
+DefaultImporter:
+  externalObjects: {}
+  userData: 
+  assetBundleName: 
+  assetBundleVariant: